Papers
Topics
Authors
Recent
Search
2000 character limit reached

Automated Simulations of Galaxy Morphology Evolution using Deep Learning and Particle Swarm Optimisation

Published 5 Apr 2019 in astro-ph.IM and astro-ph.GA | (1904.02906v1)

Abstract: The formation of Hoag-type galaxies with central spheroidal galaxies and outer stellar rings has yet to be understood in astronomy. We consider that these unique objects were formed from the past interaction between elliptical galaxies and gas-rich dwarf galaxies. We have modelled this potential formation process through simulation. These numerical simulations are a means of investigating this formation hypothesis, however the parameter space to be explored for these simulations is vast. Through the application of machine learning and computational science, we implement a new two-fold method to find the best model parameters for stellar rings in the simulations. First, test particle simulations are run to find a possible range of parameters for which stellar rings can be formed around elliptical galaxies (i.e. Hoag-type galaxies). A novel combination of particle swarm optimisation and Siamese neural networks has been implemented to perform the search over the parameter space and test the level of consistency between observations and simulations for numerous models. Upon the success of this initial step, we subsequently run full chemodynamical simulations for the derived range of model parameters in order to verify the output of the test particle simulations. We successfully find parameter sets at which stellar rings can be formed from the interaction between a gas-rich dwarf galaxy and a central elliptical galaxy. This is evidence that supports our hypothesis about the formation process of Hoag-type galaxies. In addition, this suggests that our new two-fold method has been successfully implemented in this problem search-space and can be investigated further in future applications. ~

Citations (1)

Summary

  • A novel method combining deep learning and Particle Swarm Optimization is presented to automate the parameter search for galaxy morphology evolution simulations.
  • The research successfully identified simulation parameters that reproduce Hoag-type galaxy features, which were validated by complex chemodynamical simulations, supporting the proposed interaction hypothesis.
  • The automated methodology enables astronomers to study galaxy morphology evolution efficiently and holds potential for broad application in other scientific optimization tasks.

Automated Simulations of Galaxy Morphology Evolution Using Deep Learning and Particle Swarm Optimization

This paper presents an investigation into the morphological evolution of Hoag-type galaxies utilizing a novel combination of deep learning and Particle Swarm Optimization (PSO). Hoag-type galaxies are characterized by a spheroidal core surrounded by a detached stellar ring, resembling the peculiar Hoag's Object. The research endeavors to address the complexities associated with such formations, hypothesized to result from interactions between elliptical and gas-rich dwarf galaxies.

Methodology

The study employs a two-fold method involving test particle simulations and full chemodynamical simulations. Initially, large test particle simulations are conducted to constrain the parameter space linked to stellar ring formation. The approach links a Siamese neural network to idiomatically assist with similarity measurement across simulation outputs and observed galaxies, operationalizing the PSO to traverse the parameter space effectively.

The PSO operates by leveraging both individual and global knowledge to explore multi-dimensional parameterized simulations, comprising initial position and velocity components of the interacting galaxies. This process facilitates the discovery of parameter sets conducive to genuine Hoag-type galaxy formation, significantly reducing the high computational expense traditionally linked to extensive parameter space exploration.

Results

The research successfully identifies several parameter sets wherein simulated outputs closely resemble Hoag-type galaxy features. Validations via complex chemodynamical simulations further substantiate these findings, endorsing the interaction hypothesis proposed. Results demonstrate a successful alignment between simulated morphology and observable characteristics, with the Siamese network model proving effective in distinguishing Hoag-type formations due to its perceived similarity learning capabilities.

Discussion

The proposed combination of PSO with a Siamese network for the task of evaluating simulation outputs is particularly notable, given the absence of similar methodologies in existing literature. While the PSO ensures a focus on optimal solutions within a broader parameter space, the neural network provides a quantitative basis to assess morphological outcomes.

Discussions within the paper acknowledge potential issues around local minima confinement and overfitting of the neural network. Suggestions for enhancing results involve exploring a larger parameter space and introducing alternative optimization techniques, like deep reinforcement learning.

Implications and Future Work

By automating the morphological galaxy classification, this methodology creates new opportunities for astronomers to better understand galaxy evolution without necessitating exhaustive manual analysis—a significant practical implication given the limitations of current theoretical models in accounting for galaxy formation complexities. Additionally, this method's generalization potential suggests applicability to other domains requiring similar optimization frameworks.

Future endeavors might involve expanding the parameter space to encompass additional dimensions or experimenting with various optimization techniques to further exploit the method's efficiency and robustness. With the advent of contemporary astronomical instruments providing voluminous data, leveraging machine learning for data analysis such as this emerges as a critical frontier in digital astronomy.

Overall, this paper contributes to galaxy morphology research through a sophisticated strategical approach in parameter search optimization, presenting a robust avenue for further exploration into astronomical phenomena via computational advances.

Paper to Video (Beta)

No one has generated a video about this paper yet.

Whiteboard

No one has generated a whiteboard explanation for this paper yet.

Open Problems

We haven't generated a list of open problems mentioned in this paper yet.

Continue Learning

We haven't generated follow-up questions for this paper yet.

Collections

Sign up for free to add this paper to one or more collections.